sarfchat 2.0.0__tar.gz

sarfchat-2.0.0/.github/workflows/ci.yml

@@ -0,0 +1,37 @@
+name: CI (sarfchat)
+
+on:
+  push:
+    branches: ["main"]
+  pull_request:
+    branches: ["main"]
+
+env:
+  TARKIB_OPENITI_ONLY: "1"
+  CUDA_VISIBLE_DEVICES: ""
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+
+      - name: Set up Python 3.12
+        run: uv python install 3.12
+
+      - name: Install CPU PyTorch and package dependencies
+        run: |
+          uv venv --python 3.12 .ci-venv
+          uv pip install --python .ci-venv/bin/python --index-url https://download.pytorch.org/whl/cpu torch
+          uv pip install --python .ci-venv/bin/python --no-deps -e .
+          uv pip install --python .ci-venv/bin/python numpy scipy scikit-learn tokenizers safetensors huggingface-hub pyarrow pytest
+
+      - name: Run offline CPU unit test suite
+        run: .ci-venv/bin/pytest -q
+
+      - name: Build wheel and sdist
+        run: uv build

sarfchat-2.0.0/.github/workflows/mirror-codeberg.yml

@@ -0,0 +1,27 @@
+name: Sovereign Mirror to Codeberg (sarfchat)
+
+on:
+  push:
+    branches: ["main"]
+    tags: ["v*"]
+  workflow_dispatch:
+
+jobs:
+  mirror:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout full git history
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Push mirror to Codeberg (maqayis/sarfchat)
+        env:
+          CODEBERG_TOKEN: ${{ secrets.CODEBERG_TOKEN }}
+        run: |
+          if [ -z "$CODEBERG_TOKEN" ]; then
+            echo "CODEBERG_TOKEN secret is not set; skipping mirror push."
+            exit 0
+          fi
+          git remote add codeberg "https://maqayis:${CODEBERG_TOKEN}@codeberg.org/maqayis/sarfchat.git" || true
+          git push --mirror codeberg

sarfchat-2.0.0/.github/workflows/publish.yml

@@ -0,0 +1,32 @@
+name: Publish to PyPI (sarfchat)
+
+on:
+  release:
+    types: ["published"]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  id-token: write
+
+env:
+  TARKIB_OPENITI_ONLY: "1"
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+
+      - name: Set up Python 3.12
+        run: uv python install 3.12
+
+      - name: Build wheel and sdist
+        run: uv build
+
+      - name: Publish package distributions to PyPI via OIDC Trusted Publisher
+        uses: pypa/gh-action-pypi-publish@release/v1

sarfchat-2.0.0/.gitignore

@@ -0,0 +1,23 @@
+# Byte-compiled / caches
+__pycache__/
+*.py[cod]
+*$py.class
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+
+# Virtual environments & build staging (`dist/` release wheels/sdists are tracked)
+.venv/
+venv/
+build/
+*.egg-info/
+
+# Hydrated checkpoints & corpus artifacts (fetched via ./fetch-data.sh & verified by data-manifest.tsv)
+checkpoints/
+data/
+!data-manifest.tsv
+
+# Local environment files
+.env
+.env.*
+!.env.example

sarfchat-2.0.0/CITATION.cff

@@ -0,0 +1,25 @@
+cff-version: 1.2.0
+message: "If you use SarfChat-v2 in your research or software, please cite it using the metadata below."
+title: "SarfChat-v2: Certified Byte-Lossless Root×Pattern Tokenization, Grassmannian Value Routing, and Iso-FLOP Bits-Per-Byte Scaling Laws for Arabic Foundation Models"
+version: "2.0.0"
+date-released: "2026-11-07"
+license: "Apache-2.0"
+url: "https://maqayis.org"
+repository-code: "https://github.com/maqayis/sarfchat"
+authors:
+  - family-names: "Abdo"
+    given-names: "Salah"
+    email: "Salah.Abdo.Tech@gmail.com"
+abstract: >-
+  Algebraic Semitic Root×Pattern Transformer (d16, 261.75M total / 221.79M active matmul parameters)
+  trained from scratch on 5.10B tokens (33.63 GB UTF-8) to val_bpb = 0.7353 (0.6976 on unseen OOV-root
+  passages; 1.8861 localized OOV-word bpb), paired with certified 100%-byte-lossless Sarf-BPE,
+  Grassmannian multi-asl fiber value routing, a 19-run Iso-FLOP scaling and causal permutation harness,
+  and a 23.8-microsecond GPU Tarkib morpho-phonotactic shield achieving 100.0% legality (40/40 and 200/200).
+keywords:
+  - "Arabic foundation models"
+  - "root-pattern morphology"
+  - "Grassmannian manifold"
+  - "bits-per-byte scaling laws"
+  - "Maqayis"
+  - "sarfchat"

sarfchat-2.0.0/LICENSE

@@ -0,0 +1,176 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the Work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS

sarfchat-2.0.0/NOTICE

@@ -0,0 +1,19 @@
+SarfChat-v2 (صرف-شات) v2.0.0 — Maqāyīs (مَقاييس) Open-Source Ecosystem
+Copyright 2026 Salah Abdo and successor entity
+
+Software, Model Architecture, Codec, & Evaluation Harness (`src/`, `scripts/`, `tests/`):
+  Licensed under the Apache License, Version 2.0 (Apache-2.0).
+  You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0
+
+Research Paper, Figures, & Empirical Scaling-Law / Fortification Ledgers (`paper/`, `results/`):
+  Licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
+  https://creativecommons.org/licenses/by/4.0/
+
+Clean-Room Provenance & Architectural Acknowledgements:
+  - Morphological, phonotactic, and Grassmannian multi-aṣl priors (`sarfchat_priors_v1.npz` &
+    `sarfchat_priors_v2.npz`) are compiled cleanly via `tarkib>=2.0.0` (`TARKIB_OPENITI_ONLY=1`)
+    strictly from pre-modern public-domain classical Arabic lexicons (170–1205 AH via OpenITI).
+  - Baseline ResFormer / `MuonAdamW` trunk conventions draw architectural inspiration from
+    Andrej Karpathy's public `nanochat` reference (MIT License), extended with certified
+    byte-lossless `Ṣarf-BPE`, `GrassmannFiberValueRouter`, `TarkibSubwordInjector`, single-pass
+    `_FusedChunkedSoftcapLinearCE`, and the `<24 μs` GPU `Tarkīb` morpho-phonotactic shield.

sarfchat-2.0.0/PKG-INFO

@@ -0,0 +1,115 @@
+Metadata-Version: 2.5
+Name: sarfchat
+Version: 2.0.0
+Summary: SarfChat-v2 (صرف-شات): Algebraic Semitic Root×Pattern Transformer, Certified Byte-Lossless Sarf-BPE, & Iso-FLOP Scaling-Law Harness (Maqayis)
+Project-URL: Homepage, https://maqayis.org
+Project-URL: Repository, https://github.com/maqayis/sarfchat
+Project-URL: Mirror, https://codeberg.org/maqayis/sarfchat
+Author-email: Salah Abdo <Salah.Abdo.Tech@gmail.com>
+License: Apache-2.0
+License-File: LICENSE
+License-File: NOTICE
+Keywords: arabic,grassmannian,maqayis,morphology,sarfchat,scaling-laws,transformer
+Classifier: Development Status :: 5 - Production/Stable
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Natural Language :: Arabic
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Requires-Python: >=3.11
+Requires-Dist: huggingface-hub>=0.26.0
+Requires-Dist: numpy>=1.26.0
+Requires-Dist: pyarrow>=15.0.0
+Requires-Dist: safetensors>=0.4.5
+Requires-Dist: scikit-learn>=1.4.0
+Requires-Dist: scipy>=1.12.0
+Requires-Dist: tarkib>=2.0.0
+Requires-Dist: tokenizers>=0.20.0
+Requires-Dist: torch>=2.4.0
+Provides-Extra: data
+Requires-Dist: datasets>=3.0.0; extra == 'data'
+Provides-Extra: rag
+Requires-Dist: tayf>=1.0.0; extra == 'rag'
+Description-Content-Type: text/markdown
+
+# SarfChat-v2 (`صرف-شات`) v2.0.0 — المحوِّل الصرفي–الغراسماني التأسيسي وقوانين التوسّع التوليدي
+
+<div align="center">
+
+**Algebraic Semitic Root×Pattern Foundation Transformer (`d16`, `261.75M` Params, `5.10B` Tokens), Certified `100%`-Byte-Lossless `Ṣarf-BPE`, `Grassmann-VE` Routing, & `23.8 μs` GPU `Tarkīb` Shield**  
+Part of the **Maqāyīs (`مَقاييس`)** Open-Source Ecosystem · [maqayis.org](https://maqayis.org) · [github.com/maqayis/sarfchat](https://github.com/maqayis/sarfchat) · [codeberg.org/maqayis/sarfchat](https://codeberg.org/maqayis/sarfchat)
+
+[![License: Apache-2.0](https://img.shields.io/badge/Code-Apache--2.0-blue.svg)](LICENSE)
+[![Paper & Results: CC BY 4.0](https://img.shields.io/badge/Paper_%26_Results-CC_BY_4.0-246B45.svg)](NOTICE)
+[![Python 3.11+](https://img.shields.io/badge/Python-3.11%2B-3776AB.svg)](pyproject.toml)
+
+</div>
+
+---
+
+## نظرة عامّة | Overview
+
+**`SarfChat-v2` (`صرف-شات v2.0.0`)** هو المحوِّل التوليدي التأسيسي ومختبر قوانين التوسّع متساوية الحوسبة (`Iso-FLOP Scaling Laws`) في منظومة **مَقاييس (`Maqāyīs`)** من تأليف وتطوير **صلاح عبده (`Salah Abdo`)**. بدلاً من الاكتفاء بترقيع النماذج الأجنبية من الخارج أو استنزاف ميزانية المعلمات في جداول التقطيع الإحصائي الصمّاء (`Byte-BPE`)، يدمج `SarfChat-v2` جبر الجذر والوزن الريماني وألياف غراسمان المتعدّدة الأصول (`Gr(k_R, 256)`) والسمات النطقية لسيبويه في صميم طبقات المحوِّل ذاته.
+
+**`SarfChat-v2` (`v2.0.0`)** replaces monolithic subword memorization tables with **asymmetric morphological-Grassmannian inductive biases**:
+1. **Certified `100.0%`-Byte-Lossless `Ṣarf-BPE` & `FactoredSarfDAGHead` (`Exp S1`):** Folds `65.67%` of running Arabic words (`36,186 / 55,103`) and `100.0%` (`534 / 534`) of unseen OOV-root words into a single algebraic frame `(p, C1, C2, C3, w, e)`, eliminating OOV radical fracture (`2.41 -> 1.00` steps/word) and saving `1.22×` quadratic attention FLOPs while guaranteeing `100.000%` UTF-8 byte round-trip identity.
+2. **Dynamic `Grassmann-VE` Routing (`GrassmannFiberValueRouter`, `Exp S2–S3`):** Projects hidden states onto Ibn Fāris's orthonormal multi-*aṣl* sense bundles `Gr(k_R, 256)` across `L/2` alternating layers, compressing deep value tables by **`-93.6%`** (`75.50M -> 4.83M` at `d12`; `134.22M -> 9.54M` at `d16`).
+3. **19-Run Iso-FLOP Scaling & Causal Permutation Campaign (`Exp S2–S3`):** Breaks the multi-epoch flat-BPE memorization wall (`1.0962 -> 0.9716 bpb`, **`-0.1246 bpb`** at `d12`) and causally separates low-rank value regularization on seen words (`token_perm` / `root_perm` / `true_tarkib` within `±0.003 bpb` overall) from authentic sub-root zero-shot generalization on unseen OOV-root target words (**`2.6677 bpb`** vs `2.7009` `token_perm`, `2.7007` `root_perm`, and `2.8620` flat BPE; paired `t = -5.94, p = 2.83×10⁻⁹` across `97` roots).
+4. **Asymmetric Flagship `SarfChat-v2` (`d16`, `261.75M` Params, `5.10B` Tokens, `Exp S4–S5`):** Pairs `TarkibSubwordInjector` (`23,253 / 32,768` root-grounded subwords = `70.96%`, `29.49%` interface share vs `64.52%` flat equivalent) with single-pass `_FusedChunkedSoftcapLinearCE` (`8.59 GB -> 0.54 GB` logit VRAM, **`-93.7%`**, `105.0k tok/s`, **`58.3% MFU`**) and paragraph-level OOV surgical isolation (`1.4% -> 57.7%` long-book retention = **`41.2×`**, `0.0%` root leakage).
+5. **`23.8 μs` GPU `Tarkīb` Morpho-Phonotactic Shield & `Tayf` Neuro-Symbolic Prism (`Exp S6–S7`):** Lifts morpho-phonotactic & ontological legality to **`100.0%`** (`40/40` core suite: `97.5% -> 100.0%`; `200/200` extended suite: `98.0% / 96.5% -> 100.0%`) in `23.82 μs` per step, and lowers grounded answer entropy from `1.238 -> 0.673 bpb` (`95.0%` homonym trap survival) when paired with Zero-LLM [`maqayis/tayf`](https://github.com/maqayis/tayf).
+
+---
+
+## النتائج المرجعية | Headline `SarfChat-v2` Metrics
+
+| Metric / Stage | Specification / Split | Value |
+| :--- | :--- | :--- |
+| **Architecture (`sarfchat_v2`)** | `d16` (`L=16, d=1024, n_q=16, n_kv=8, T=2048, V=32,768`) | **`261,751,467`** total / **`221,788,952`** active matmul |
+| **Interface Parameter Tax** | `SarfChat-v2` (`77.20M`) vs Flat Equivalent (`335.54M`) | **`29.49%`** vs `64.52%` (`-258.3M` params saved) |
+| **Pretraining Scale (Stages 1 + 2)** | `WSD Plateau` (`4.40B`) + `WSD Cooldown` (`0.70B`) | **`5,095,292,928` tokens** (**`33.63 GB`** UTF-8, `8.83e18` FLOPs) |
+| **Overall Validation `val_bpb`** | Held-out multi-register `val_v2` (`0.8157` plateau -> cooldown) | **`0.7353 bpb`** (`0.73529`, `-0.0804` cooldown drop) |
+| **Register Breakdown (`val_bpb`)** | `msa_edu` / `msa_wiki_books` / `classical_turath` | **`0.7139`** / **`0.7321`** / **`0.7598 bpb`** |
+| **Zero-Shot Unseen OOV-Root `bpb`** | `511` OOV-root docs (`100` strictly held-out roots, `0%` leakage) | **`0.6976 bpb`** (`0.69757` doc) · **`1.8861 bpb`** (`643` OOV words) |
+| **Stage 3 Arabic Chat SFT** | `46,858` train / `600` val multi-turn dialogues (`342` steps) | `sft_val_loss`: `4.0122 ->` **`2.4743`** (`PPL = 11.87`) |
+| **GPU `Tarkīb` Guard (`23.82 μs`)** | `40`-pair core suite (`base`) · `200`-pair fortification (`base` / `sft`) | `97.5% ->` **`100.0%`** (`40/40`) · `98.0% / 96.5% ->` **`100.0%`** (`200/200`) |
+| **Neuro-Symbolic `Tayf + SarfChat-v2`** | `20`-query adversarial homonym-trap grounded QA (`Exp S7-B`) | `1.238 ->` **`0.673 bpb`** (`-45.7%`) · **`95.0%`** (`19/20`) trap survival |
+
+> **Book `v2.0` Sync Note:** *The Geometry of the Arabic Tongue* (`Book v2.0`, [`maqayis/tarkib`](https://github.com/maqayis/tarkib)) Ch. 21 reports the foundational `SarfChat-v1` `d12` Iso-FLOP campaign and early `d16` specification, while Part X (Ch. 22–23) and this repository (`v2.0.0`) present the completed `5.10B`-token 3-stage `d16` WSD cooldown + Chat SFT campaign (`val_bpb = 0.7353`), `19`-run causal permutation matrix, and `200`-pair GPU guard audit.
+
+---
+
+## التثبيت والتشغيل السريع | Installation & Quickstart
+
+```bash
+pip install sarfchat
+```
+
+### 1. Offline CPU Unit Tests (`<200 MB` RAM, Zero Downloads Required)
+```bash
+pytest -q
+```
+
+### 2. Hydrate Checkpoints & Tokenizers (SHA-256 Verified via `data-manifest.tsv`)
+```bash
+./fetch-data.sh data   # corpus splits, bpe16k/bpe32k tokenizers, and Tarkīb priors (~499 MB)
+./fetch-data.sh runs   # all 8 safetensors checkpoints incl. sarfchat_v2_d16_base & _sft (~2.96 GB)
+```
+
+### 3. Launch Interactive RTL Web Arena or Terminal Chat
+```bash
+sarfchat-arena --host 127.0.0.1 --port 8787
+sarfchat-chat --checkpoint checkpoints/sarfchat_v2_d16_sft
+```
+
+See [`WEIGHTS_AND_REPRODUCIBILITY.md`](WEIGHTS_AND_REPRODUCIBILITY.md) for complete `SHA-256` checkpoint manifests, model cards, and exact reproduction commands, and [`paper/main.pdf`](paper/main.pdf) (`paper/main.tex`) for the full Arabic monograph study.
+
+---
+
+## الترخيص وحقوق الملكية | License & Attribution
+
+- **Software & Source Code (`src/`, `scripts/`, `tests/`):** [Apache License 2.0](LICENSE)
+- **Research Paper, Figures, & Empirical Result Ledgers (`paper/`, `results/`):** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)
+- **Author:** Salah Abdo (`صلاح عبده`) `<Salah.Abdo.Tech@gmail.com>` · `Copyright 2026 Salah Abdo and successor entity` · [`NOTICE`](NOTICE) · [`CITATION.cff`](CITATION.cff)

sarfchat-2.0.0/README.md

@@ -0,0 +1,78 @@
+# SarfChat-v2 (`صرف-شات`) v2.0.0 — المحوِّل الصرفي–الغراسماني التأسيسي وقوانين التوسّع التوليدي
+
+<div align="center">
+
+**Algebraic Semitic Root×Pattern Foundation Transformer (`d16`, `261.75M` Params, `5.10B` Tokens), Certified `100%`-Byte-Lossless `Ṣarf-BPE`, `Grassmann-VE` Routing, & `23.8 μs` GPU `Tarkīb` Shield**  
+Part of the **Maqāyīs (`مَقاييس`)** Open-Source Ecosystem · [maqayis.org](https://maqayis.org) · [github.com/maqayis/sarfchat](https://github.com/maqayis/sarfchat) · [codeberg.org/maqayis/sarfchat](https://codeberg.org/maqayis/sarfchat)
+
+[![License: Apache-2.0](https://img.shields.io/badge/Code-Apache--2.0-blue.svg)](LICENSE)
+[![Paper & Results: CC BY 4.0](https://img.shields.io/badge/Paper_%26_Results-CC_BY_4.0-246B45.svg)](NOTICE)
+[![Python 3.11+](https://img.shields.io/badge/Python-3.11%2B-3776AB.svg)](pyproject.toml)
+
+</div>
+
+---
+
+## نظرة عامّة | Overview
+
+**`SarfChat-v2` (`صرف-شات v2.0.0`)** هو المحوِّل التوليدي التأسيسي ومختبر قوانين التوسّع متساوية الحوسبة (`Iso-FLOP Scaling Laws`) في منظومة **مَقاييس (`Maqāyīs`)** من تأليف وتطوير **صلاح عبده (`Salah Abdo`)**. بدلاً من الاكتفاء بترقيع النماذج الأجنبية من الخارج أو استنزاف ميزانية المعلمات في جداول التقطيع الإحصائي الصمّاء (`Byte-BPE`)، يدمج `SarfChat-v2` جبر الجذر والوزن الريماني وألياف غراسمان المتعدّدة الأصول (`Gr(k_R, 256)`) والسمات النطقية لسيبويه في صميم طبقات المحوِّل ذاته.
+
+**`SarfChat-v2` (`v2.0.0`)** replaces monolithic subword memorization tables with **asymmetric morphological-Grassmannian inductive biases**:
+1. **Certified `100.0%`-Byte-Lossless `Ṣarf-BPE` & `FactoredSarfDAGHead` (`Exp S1`):** Folds `65.67%` of running Arabic words (`36,186 / 55,103`) and `100.0%` (`534 / 534`) of unseen OOV-root words into a single algebraic frame `(p, C1, C2, C3, w, e)`, eliminating OOV radical fracture (`2.41 -> 1.00` steps/word) and saving `1.22×` quadratic attention FLOPs while guaranteeing `100.000%` UTF-8 byte round-trip identity.
+2. **Dynamic `Grassmann-VE` Routing (`GrassmannFiberValueRouter`, `Exp S2–S3`):** Projects hidden states onto Ibn Fāris's orthonormal multi-*aṣl* sense bundles `Gr(k_R, 256)` across `L/2` alternating layers, compressing deep value tables by **`-93.6%`** (`75.50M -> 4.83M` at `d12`; `134.22M -> 9.54M` at `d16`).
+3. **19-Run Iso-FLOP Scaling & Causal Permutation Campaign (`Exp S2–S3`):** Breaks the multi-epoch flat-BPE memorization wall (`1.0962 -> 0.9716 bpb`, **`-0.1246 bpb`** at `d12`) and causally separates low-rank value regularization on seen words (`token_perm` / `root_perm` / `true_tarkib` within `±0.003 bpb` overall) from authentic sub-root zero-shot generalization on unseen OOV-root target words (**`2.6677 bpb`** vs `2.7009` `token_perm`, `2.7007` `root_perm`, and `2.8620` flat BPE; paired `t = -5.94, p = 2.83×10⁻⁹` across `97` roots).
+4. **Asymmetric Flagship `SarfChat-v2` (`d16`, `261.75M` Params, `5.10B` Tokens, `Exp S4–S5`):** Pairs `TarkibSubwordInjector` (`23,253 / 32,768` root-grounded subwords = `70.96%`, `29.49%` interface share vs `64.52%` flat equivalent) with single-pass `_FusedChunkedSoftcapLinearCE` (`8.59 GB -> 0.54 GB` logit VRAM, **`-93.7%`**, `105.0k tok/s`, **`58.3% MFU`**) and paragraph-level OOV surgical isolation (`1.4% -> 57.7%` long-book retention = **`41.2×`**, `0.0%` root leakage).
+5. **`23.8 μs` GPU `Tarkīb` Morpho-Phonotactic Shield & `Tayf` Neuro-Symbolic Prism (`Exp S6–S7`):** Lifts morpho-phonotactic & ontological legality to **`100.0%`** (`40/40` core suite: `97.5% -> 100.0%`; `200/200` extended suite: `98.0% / 96.5% -> 100.0%`) in `23.82 μs` per step, and lowers grounded answer entropy from `1.238 -> 0.673 bpb` (`95.0%` homonym trap survival) when paired with Zero-LLM [`maqayis/tayf`](https://github.com/maqayis/tayf).
+
+---
+
+## النتائج المرجعية | Headline `SarfChat-v2` Metrics
+
+| Metric / Stage | Specification / Split | Value |
+| :--- | :--- | :--- |
+| **Architecture (`sarfchat_v2`)** | `d16` (`L=16, d=1024, n_q=16, n_kv=8, T=2048, V=32,768`) | **`261,751,467`** total / **`221,788,952`** active matmul |
+| **Interface Parameter Tax** | `SarfChat-v2` (`77.20M`) vs Flat Equivalent (`335.54M`) | **`29.49%`** vs `64.52%` (`-258.3M` params saved) |
+| **Pretraining Scale (Stages 1 + 2)** | `WSD Plateau` (`4.40B`) + `WSD Cooldown` (`0.70B`) | **`5,095,292,928` tokens** (**`33.63 GB`** UTF-8, `8.83e18` FLOPs) |
+| **Overall Validation `val_bpb`** | Held-out multi-register `val_v2` (`0.8157` plateau -> cooldown) | **`0.7353 bpb`** (`0.73529`, `-0.0804` cooldown drop) |
+| **Register Breakdown (`val_bpb`)** | `msa_edu` / `msa_wiki_books` / `classical_turath` | **`0.7139`** / **`0.7321`** / **`0.7598 bpb`** |
+| **Zero-Shot Unseen OOV-Root `bpb`** | `511` OOV-root docs (`100` strictly held-out roots, `0%` leakage) | **`0.6976 bpb`** (`0.69757` doc) · **`1.8861 bpb`** (`643` OOV words) |
+| **Stage 3 Arabic Chat SFT** | `46,858` train / `600` val multi-turn dialogues (`342` steps) | `sft_val_loss`: `4.0122 ->` **`2.4743`** (`PPL = 11.87`) |
+| **GPU `Tarkīb` Guard (`23.82 μs`)** | `40`-pair core suite (`base`) · `200`-pair fortification (`base` / `sft`) | `97.5% ->` **`100.0%`** (`40/40`) · `98.0% / 96.5% ->` **`100.0%`** (`200/200`) |
+| **Neuro-Symbolic `Tayf + SarfChat-v2`** | `20`-query adversarial homonym-trap grounded QA (`Exp S7-B`) | `1.238 ->` **`0.673 bpb`** (`-45.7%`) · **`95.0%`** (`19/20`) trap survival |
+
+> **Book `v2.0` Sync Note:** *The Geometry of the Arabic Tongue* (`Book v2.0`, [`maqayis/tarkib`](https://github.com/maqayis/tarkib)) Ch. 21 reports the foundational `SarfChat-v1` `d12` Iso-FLOP campaign and early `d16` specification, while Part X (Ch. 22–23) and this repository (`v2.0.0`) present the completed `5.10B`-token 3-stage `d16` WSD cooldown + Chat SFT campaign (`val_bpb = 0.7353`), `19`-run causal permutation matrix, and `200`-pair GPU guard audit.
+
+---
+
+## التثبيت والتشغيل السريع | Installation & Quickstart
+
+```bash
+pip install sarfchat
+```
+
+### 1. Offline CPU Unit Tests (`<200 MB` RAM, Zero Downloads Required)
+```bash
+pytest -q
+```
+
+### 2. Hydrate Checkpoints & Tokenizers (SHA-256 Verified via `data-manifest.tsv`)
+```bash
+./fetch-data.sh data   # corpus splits, bpe16k/bpe32k tokenizers, and Tarkīb priors (~499 MB)
+./fetch-data.sh runs   # all 8 safetensors checkpoints incl. sarfchat_v2_d16_base & _sft (~2.96 GB)
+```
+
+### 3. Launch Interactive RTL Web Arena or Terminal Chat
+```bash
+sarfchat-arena --host 127.0.0.1 --port 8787
+sarfchat-chat --checkpoint checkpoints/sarfchat_v2_d16_sft
+```
+
+See [`WEIGHTS_AND_REPRODUCIBILITY.md`](WEIGHTS_AND_REPRODUCIBILITY.md) for complete `SHA-256` checkpoint manifests, model cards, and exact reproduction commands, and [`paper/main.pdf`](paper/main.pdf) (`paper/main.tex`) for the full Arabic monograph study.
+
+---
+
+## الترخيص وحقوق الملكية | License & Attribution
+
+- **Software & Source Code (`src/`, `scripts/`, `tests/`):** [Apache License 2.0](LICENSE)
+- **Research Paper, Figures, & Empirical Result Ledgers (`paper/`, `results/`):** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)
+- **Author:** Salah Abdo (`صلاح عبده`) `<Salah.Abdo.Tech@gmail.com>` · `Copyright 2026 Salah Abdo and successor entity` · [`NOTICE`](NOTICE) · [`CITATION.cff`](CITATION.cff)